|
|
|
Teknowledge |
|
Adam Pease, John Li, Bhanu Vasireddy, Jeff Vu,
Joe Marcelino |
|
[apease | jli]@teknowledge.com |
|
February 13, 2002 |
|
|
|
|
|
Scaleable DAML-based search engine |
|
Answers to queries, not pages |
|
Experiments to quantify performance |
|
Includes translation between ontologies |
|
Includes simple inferences to broaden queries |
|
Query continuation |
|
Deployed at: (first deployed end of September) http://plucky.teknowledge.com/daml/damlquery.jsp |
|
Downloadable as a zip file for local
installation on http://reliant.teknowledge.com/DAML/ |
|
Large ontologies in DAML, DAML-KIF translator |
|
Financial ontology to Horus |
|
Financial Instrument, Financial Contract,
Currency Measure, etc |
|
Example commercial application |
|
|
|
|
|
|
|
|
|
|
|
|
|
Weekly update process |
|
Crawling |
|
Pre-processing |
|
Inference caching into tables |
|
Deployment |
|
|
|
Done via automated scripts |
|
|
|
|
|
|
|
|
|
samePropertyAs, sameClassAs, (sameIndividualAs) |
|
inverseOf |
|
Search for (?X childOf ?Y) should also return
results when content is coded as
(?Y parentOf ?X) |
|
subPropertyOf, subClassOf |
|
Generalization and specialization |
|
Search is ordered so exact matches are returned
first and broadening happens next |
|
|
|
|
Ontology references can have typos |
|
DAML is mixed with XML and HTML and all may be
malformed |
|
Sites change, are unavailable for short, long
and permanent periods |
|
Massive amounts of content – space and time
efficiency issues |
|
53,000 pages, 3.7 million DAML statements, 800MB
crawled DB size |
|
Currently using SirPac, exploring JENA |
|
|
|
|
|
Quantitative Experiments not “Existence proofs” |
|
Metrics include: search time on different
queries, inter-agent communication speed (as mediated by ASCS), correctness
of the translation, semantic complexity of source and target communication |
|
|
|
|
|
WordNet in DAML |
|
10,000 pages each with 10-100 DAML statements |
|
IEEE Standard Upper Ontology (proposal) |
|
Large, free, general purpose, formal,
peer-reviewed |
|
People are free to reuse all or part of the
content |
|
A resource to speed creation of good ontologies
and avoid reinvention |
|
~2000 DAML statements |
|
http://ontology.teknowledge.com |
|
|
|
|
|
|
Financial ontology in DAML (and built on the SUO
proposal) |
|
DAML encoded aggregation displays |
|
DAML markup of bond sites |
|
Ontology translation |
|
Crawl and index bond sites |
|
Dynamic search, translation, advice |
|
|
|
|
Use rdfs:label to give human-readable labels to
all items |
|
Use sameClassAs and samePropertyAs to relate
your ontology to others |
|
|
|
|
|
|
Fully implement distributed architecture |
|
Determine how to partition the DAML search space |
|
One ontology per agent or several? |
|
Improve speed and space efficiency |
|
Supply search and translation agents for DAML
experiment |
|
Some manual ontology mapping |
|
Semi-automated discovery of translations |
|
Deployment of client and server-based search
agents |
|
Multi-user DAML-based advice server |
|
Improved general search interface |
|
New version will be deployed as well as posted
for download every 3 months |
|
Use RuleML for ontology mapping |
|
|
|
Notes